realismo

name: realismo description: > Skill for generating photorealistic images indistinguishable from reality. Should be used when the goal is to create photos that look like they were taken by a real person with a real phone or camera — not by a professional studio and not by AI. Applies calculated human imperfections to break the "artificial perfection" that exposes AI-generated images. Relevant for: product photos, lifestyle, lookbook, social media, e-commerce, catalog, any scenario where the image must look authentic and organic.

Imperceptible Realism

Core Philosophy

Perfection is the enemy of realism.

AI images fail to look real not because of lack of quality, but because of excess quality. Real photos taken by real people carry subtle imperfection markers that our brain unconsciously recognizes as "authentic." When these markers are absent, something feels "off" — even if the viewer can't explain what.

The goal of this skill is not to generate a beautiful photo. It is to generate a photo that no one questions.

The Imperfect Authenticity Principle

Every real photo carries 3 layers of imperfection:

1. Equipment Imperfection

The camera or phone leaves its mark:

• Slight chromatic aberration at the edges
• Subtle grain/noise (especially in shadows)
• Micro-focus variation (not everything is perfectly sharp)
• JPEG compression (subtle artifacts in high-contrast areas)
• Slight lens distortion (barrel distortion from phone wide-angle)

2. Photographer Imperfection

The person taking the photo is not perfect:

• Composition almost centered, but not mathematically perfect
• Horizon slightly tilted (0.5° to 2°)
• Subject unintentionally cropped (top of head, foot, elbow)
• Captured moment between poses (not the "best moment")
• Subject distance is not the ideal studio distance

3. Environment Imperfection

The real world doesn't cooperate:

• Background objects that weren't "cleaned" (cup on table, loose hair strand)
• Mixed lighting (natural + artificial light blending)
• Unwanted reflections on surfaces (glasses, storefronts, screens)
• Shadows in slightly inconsistent directions
• Micro-movement (slight motion blur on extremities: hair, hands)

How to Apply: The 7 Levers of Realism

When building any prompt that needs to look real, inject at least 3 of the 7 levers below. Never use all of them at once — overdoing imperfection is just as artificial as perfection.

Lever 1: Real Device + Lens

State that the photo was taken with a specific device and lens combination.

Concept: The model adjusts aberration, grain, compression, and perspective to simulate the optical characteristics of that device. The lens specification is equally important — it changes how texture is compressed and rendered.

Device × Context pairing:

Lens pairing for textile work:

💡 Macro lens = obrigatório para shots de detalhe de trama (tricô, crochê, rendas). Sem ele, o modelo gera textura interpolada, não fibras reais.

Lever 2: Uncontrolled Lighting

Describe lighting that happens, not lighting that was designed.

Concept: Studio lighting (softbox, ring light) screams "professional production." Natural lighting with inconsistencies screams "real moment."

Useful elements:

• Light coming through a window, creating hard shadows on one side
• Mix of warm (tungsten) and cool (daylight) light
• Unintentional backlight (subject partially silhouetted)
• Ambient lighting (ceiling light, storefront, partial sunset)

Lever 3: Organic Composition

Describe framing that a real person would do.

Concept: Professional photographers use rule of thirds, leading lines, and calculated negative space. Real people point the phone and press the button.

Useful elements:

• Subject slightly off-center
• Something at the edge of the frame that shouldn't be there
• Angle slightly above or below eye level
• Slight digital zoom (instead of optimal focal length)

Lever 4: Surface Texture

Describe textures with natural defects.

Concept: AI tends to generate homogeneous, uniform surfaces. The real world has texture variation on everything.

Useful elements:

• Skin with visible pores, slight oiliness, sun marks
• Fabric with micro-wrinkles from use, not "freshly ironed"
• Surfaces with subtle dust, fingerprints, microscopic scratches
• Hair with loose strands and flyaways (not perfectly styled)

Lever 5: Moment, Not Pose

Describe action between moments, not at the perfect moment.

Concept: Most real photos capture the instant before or after the "ideal moment." It's the "almost-smile," the hand still in motion, the gaze that hasn't found the camera yet.

Useful elements:

• Mid-action (in the middle of a gesture, not at the final position)
• Gaze directed at something off-frame (not at the camera)
• Ambiguous facial expression (not the perfect smile)
• Relaxed and asymmetrical posture

Lever 6: Imperfect Depth

Describe focus that almost got it right.

Concept: Phone autofocus and consumer cameras sometimes focus on the wrong point — on the background instead of the subject, on the shoulder instead of the eye. A microscopically wrong focus is the strongest marker of a real photo.

Useful elements:

• Focus on chest/shoulder instead of eyes
• Foreground slightly soft, background sharp (opposite of expected)
• One hand sharp, the other slightly out of focus
• Slightly irregular bokeh (not perfectly circular)

Lever 7: Capture Artifacts

Describe marks from the real capture process.

Concept: Every real photo carries evidence that it was captured, processed, and compressed by a real technology chain.

Useful elements:

• Subtle film grain or digital noise in shadows
• Slight purple fringing in high-contrast areas
• Visible JPEG compression in sky gradients
• Uncorrected white balance (slightly too warm or too cool)
• Natural lens vignetting

Lever 8: Grain Injection (corrige AI Softness)

Conceito: Imagens geradas por IA às vezes sofrem de "AI Softness" — uma nitidez artificial ou suavidade excessiva nas fibras e superfícies que delata a origem digital. A Grain Injection ancora a imagem na realidade quebrando essa homogeneidade.

Quando usar: após gerar uma imagem de tecido texturizado que parece "plástico" ou "pintado" nas fibras.

🔑 Primeira linha de defesa: antes de usar Grain Injection, tente subir o thinking_level de MINIMAL para MEDIUM (ou HIGH para texturas muito complexas como Aran, crochê, renda). O Thinking Mode executa Plan → Evaluate → Improve e frequentemente resolve o AI Softness sem necessidade de pós-processamento.

No prompt (quando thinking upgrade não resolver):

Subtle monochromatic grain (1-3%), natural fiber micro-texture visible, 
no AI smoothing on fabric surface, yarn filaments individually distinct.

Em pós-processamento (Photoshop/Lightroom):

1. Aplicar filtro de ruído monocromático (1–3%) sobre a imagem
2. Usar filtro "Mínimo" na máscara de camada para esconder frangias brancas/pretas nas bordas da roupa
3. Resultado: textura de fibra âncora na percepção de realidade

⚠️ Não exagere. Acima de 4% de grain o efeito se torna visível como artifício — especialmente em pele.

Anti-Patterns: What NOT to Do

These practices destroy realism and turn the image into "obvious AI":

Calibrating Intensity

Not every image needs the same level of "amateurism." Calibrate according to context:

Level 1 — Casual Authentic (social media, lifestyle)

Use 5-7 levers at high intensity. Should look like: selfie, friend's photo, Instagram story.

Level 2 — Semi-Professional (e-commerce, catalog)

Use 3-4 levers at moderate intensity. Should look like: photo taken with a good phone by someone who understands a bit of photography, but isn't professional. The type of photo an indie brand would post.

Level 3 — Natural Professional (editorial, lookbook)

Use 2-3 levers at subtle intensity. Should look like: independent magazine editorial, fashion blog. Well-photographed, but not over-produced.

Level by Shot Type (quick reference)

💡 Regra rápida: Ghost mannequin e ML compliance = menos realismo (aparência de catálogo profissional). Lifestyle e Shopee hero = mais realismo (parece foto real de pessoa real).

Mental Validation Checklist

Before finalizing a prompt, run through this checklist:

• Does the photo have at least one thing "wrong"? (If everything is perfect, it's AI)
• Is the lighting mixed or natural? (If it's studio lighting, it's AI)
• Is the subject between moments? (If they're in the perfect pose, it's AI)
• Is there something in the background that "shouldn't be there"? (If the background is clean, it's AI)
• Is the focus microscopically off? (If everything is 100% sharp, it's AI)
• Is the composition organic? (If it's mathematically centered, it's AI)
• Did I avoid forced quality tags? (8K, masterpiece, ultra HD)

Integration with Other Workflows

This skill is a conceptual layer that applies on top of any other workflow or prompt. It does not replace the description of the subject, clothing, or setting — it adds the authenticity layer that makes the difference between "impressive AI" and "real photo."

Recommended flow:

1. Build the base prompt (subject, clothing, setting, action)
2. Apply the realism levers from this skill
3. Remove any term that breaks authenticity (checklist above)
4. Review: "If I saw this photo on Instagram, would I be suspicious?"